Labor Supply Responses of Italian Women to. Minimum Income Policies

Transcription

1 Labor Supply Responses of Italian Women to Minimum Income Policies Anna Laura Mancini May 9, 2008 Abstract Minimum income policies are policies aimed at guarantee all citizens with a minimum level of income and at ghting social exclusion typically associated with extreme poverty. Theoretically, their main shortcoming is the disincentive e ect on labour market participation they could generate in the bottom part of income distribution, due to the high e ective marginal tax rate they impose around the threshold level. This paper employs a structural labor supply model under discrete choices to test the existence and the magnitude of this disincentive e ect on Italian female labor supply. Our empirical results show that family structure is crucial in determining the existence of a disincentive e ect: only married women experience it, while single women participation rates increase under all possible minimum income schemes. The magnitude of both the positive and the negative e ect depend on the policy design. Child - Collegio Carlo Alberto and University of Turin 1

2 Introduction The idea of guaranteeing every citizen with a minimum level of income goes back in the history of philosophical, political and economic thought (among others, Friedman, Tobin and Van Parijs) and, in the recent years, regained the center of attention of the European political agenda 1. On one side, in favour of the minimum income idea there are motivations of redistribution, e ciency and cost-e ectiveness. On the other side, its main theoretical shortcomings are the disincentive e ect to labour market participation at the bottom end of the income distribution, due to the high e ective marginal tax rate imposed near the threshold level, and the level of taxation it would require in order to nance it. In this paper, we focus our attention on the rst critical argument against minimum income policies: the labor disincentive e ect. The problem arises from the fact that, for a low wage individual, it could be more convenient, in the shortrun, to remain out of or even to leave the labour market in order to receive the social transfer. Looking at the long run, minimum income policies could, in principle, have the undesirable e ect of creating welfare dependent families by preventing some individuals from participating in the labour market. Moreover, due to an income e ect, individuals have no incentive to work if they can get for free the same amount of money by the State. We contribute to the existing literature by testing empirically the existence and the magnitude of this labor disincentive e ect. Therefore, we investigate how labor participation would react to the introduction, in the Italian welfare system, of a basic minimum income scheme. We focus our attention on female labor supply because it is likely that the labor disincentive e ect would concern primarily women labor decisions, due to their higher exibility (as shown, for Italy, by Colombino and Del Boca 1990 and, more in general, by Laroque and 2

3 Salanie 2002 ). We choose Italy because it does not have a minimum income policy and there is wide consensus among Italian economists and sociologists that it would strongly need one to replace its highly fragmentary and workrelated actual welfare system (Sacchi 2005). The rest of the paper is as follows. Section 1 brie y introduces minimum income scheme and reviews the existing literature. Section 2 describes the 2002 Italian tax and bene t system. Section 3 illustrates the data used and the main descriptive statistics of the selected sample. Section 4 lays out the labour supply model and section 5 presents the estimation results. Section 6 describes the policy design and reports policy simulation results. Section 7 concludes. I Minimum Income Policies State intervention is primarily aimed at guaranteeing all citizens with a minimum standard of living by means of both money transfers and in-kind services. It could be divided in two main categories according to the criterion used to distribute those bene ts (Targetti Lenti 2000). The rst category uses a universal selection principle: the role of the State is mainly redistributive and each citizen derives bene ts regardless of individual particulars. Thinking about money transfers, proposals like social dividend or citizenship transfer belong to this type (see, for example, Van Parijs and Vanderborght 2006). They are universal and unconditional transfers to all citizens not included in the tax base; as an example, with a constant tax rate t, the relation between disposable income Y post, social transfer G and taxable income Y pre is: Y post = G + (1 t)y pre (1) The public health care system and the state education system can be seen as 3

4 examples of in-kind services belonging to this class of intervention. The second category is based on a selective principle: the role of the State is mainly residual and the bene ts are targeted to speci c groups of citizens, like working people, or are means-tested. Typical examples belonging to this category are minimum pensions, minimum income transfers and, in general, all social policies aimed at ghting poverty. They are realized mainly through a negative income tax (NIT) scheme where individuals with a pretax income Y pre higher than a certain threshold Y pay taxes on the exceeding part according to the country s tax rate structure. Instead, those who have an income Y pre below Y pay no taxes and receive a money transfer G from the State to increase their disposable income up to Y. Considering again a tax system with a constant tax rate t, the negative income tax works as follows: 8 >< Y pre + G with G = Y Y pre if Y pre <Y Y post = (2) >: Y + (1 t)(y pre Y ) if Y pre >Y NIT implicitly imposes a very high marginal tax rate on incomes around the threshold Y. A possible solution to avoid this problem is to weight pre-tax income Y pre by a reduction rate t 1 (lower than 1 2 ) and, consequently, to assign to individuals with t 1 (Y pre ) lower than Y a transfer G, not included in the tax base, and make them pay taxes on income higher than Y t 1 according to the country s tax structure (Fortin et al. 1993). The system works as follows: 8 >< Y pre + G with G = Y t 1 (Y pre ) if Y pre < Y t Y post = 1 (3) >: (1 t 2 )(Y pre Y ) + Y t 1 if Y pre > Y t 1 Theoretically, all these transfer schemes induce a labor disincentive e ect for those near the threshold Y. Minimum income schemes, whose main goal is to guarantee all citizens with a 4

5 minimum level of income, can belong to both categories. Universal basic income and universal basic wealth schemes are examples of the rst category, since they provide an unconditional transfer to all citizens. Workfare and participation basic income, instead, match the second category, since the transfer is conditioned to individual characteristics. Participation basic income is probably the more commonly used minimum income scheme. It is made up of two parts: a bene t scheme, to supply individuals whose income is below a certain threshold with a money transfer, and a participation program the individual has to carry out in order not to lose the monetary side. All activities included in the participation scheme are aimed at helping the individual back in to the labour market and into society on a long term perspective. In workfare schemes, instead, the money transfer is conditioned to a minimum amount of working hours. Workfare and participation systems could be designed using a NIT scheme. Most European countries already have some sort of minimum income policy, mainly modelled as a participation scheme, while Italy does not have any. In 1998, a rst experiment, the reddito minimo d inserimento, was carried out on 39 municipalities over a period of two years to test the nancial and organizational feasibility of a national minimum income scheme. In 2001, without waiting the evaluation results 3, the nancial law extended the experimentation period by two more years and increased the number of cities involved up to 306. In the minimum income experiment was declared over and in principle the reddito di ultima istanza was created in its place, but, in practice, it never actually happened. The reddito minimo d inserimento was a participation basic income mainly modelled on those already up and running in other European countries. Every city had to manage the social side autonomously, while the economic side was mainly nanced and established by the central government through the setting up of the eligibility rules and the income threshold, equal 4 5

6 to 282 euros per month (equivalent individual income was computed using the ISE scale 5 ). The common solution to avoid the disincentive problem is to exclude part of the labour earnings from the income considered to establish program eligibility. For example, in France only 50% 6 of individual earnings enter into the income taken into consideration (Gurgand and Margolis 2005), while in Portugal the percentage increases up to 70% (Rodrigues 2003). In the reddito minimo d inserimento, it was 75%. Not many of the existing empirical studies on the relation between labor supply and di erent tax-bene t structures use a discrete choice approach. Aaberge et al examine the welfare and labor supply e ects for Italian married couples of replacing the Italian tax system by three alternative schemes: a at tax, a negative income tax and a work fare system. Whatever the reform, labor supply of women in the poorest decile of the population always increases. They explain this apparently counterintuitive result, opposite to the labor disincentive e ect hypothesis, using the own- and cross-wage elasticities associated to the di erent income groups and the quantity constraints on the hours choice. Bargain and Orsini 2006 study the impact of two di erent in-work transfers in three di erent European countries, namely France, Germany and Finland, exploiting the di erences in their existing tax-bene ts systems and in the distributions of income and wages. When family transfers are considered, they nd that married women labor supply decreases, mainly due to the scal existing systems that penalize second earners, while single women labor supply increases. When individual transfers are considered, instead, the labor supply of all women increases. Blundell et al consider the impact of working families tax credit on hours and participation in UK. They nd that participation among single women increases, while married women labor supply decreases. 6

7 Very few empirical studies, to our knowledge, focus their attention on minimum income schemes. Gouveia and Rodrigues 2002, for example, study the e ect of the Portuguese Minimum income program on income distribution and on government expenditures. Gurgand and Margolis 2005 analyze the monetary work incentive faced by the recipients of the minimum income program in France, i.e. the gap between the labor market income they can earn and the welfare provision they can get. They nd that almost all welfare bene ciaries would gain from being employed rather than to stay on welfare but the size of these gains is small and it is sensitive to the way in which the authors construct the gains, in particular for single mothers. Our contribution to the existing literature is to determine the existence and the magnitude of the labor disincentive e ect associated speci cally to public transfers, since it is one of the two main theoretical argument against minimum income policies. We concentrate on a single country, Italy, to be sure that our results will not depend on di erences in the tax structure, and we use a structural family labor supply model among a set of discrete choice model to account for the fact that individuals face constraints on their possible working hours ( Dickens and Lundberg 1993, Van Soest 1995 ). In particular, we investigate what would be the labor disincentive e ect on Italian labour participation if a social transfer like the one tested between 1998 and 2003 took place. The focus on female employment is well documented in the literature, as shown by Laroque and Salanie For Italy in particular, Colombino and Del Boca 1990 and Aaberge et al show that female labour supply exibility and responsiveness are much higher with respect to male labour supply. Therefore, it is likely that the labor disincentive e ect would concern primarily female labor participation. Italian female employment rate is among the lowest in Europe (in 2006 it was equal to 46,6% while male employment rate was 70,7%) and far 7

8 below the 60% established by Lisbon target. The potential detrimental e ect on female labor supply should, then, be a major concern when considering the feasibility of an Italian minimum income scheme. II The 2002 Italian Tax-Bene t System The progressive income tax, IRPEF (Imposta sul reddito delle persone siche), represents the main source of revenue of the Italian tax system. The unit determining the taxable income is the individual, while family composition a ects the tax liability by means of tax credits for dependent spouse and dependent children. The tax base is mainly given by earnings (from employment, selfemployment or rms) and income from real estate. Income from nancial assets is normally taxed separately. In 2002 the tax schedule was made by 5 brackets with marginal rates going from 18% to 45%, as shown in table 2.1. Final tax liability depends on a system of tax credits, generally decreasing with family income, linked to the source of earned income and to dependent relatives (table 2.2). Tax credit for earned income depends on whether the individual is employed, self-employed or an entrepreneur and decreases with taxable income. In 2002, for employed individuals, it varied from a maximum of 1.146,53 euros, for gross earnings lower than euros, to a minimum of 51,65 euros, for gross earnings higher than euros. For the self-employed and entrepreneurs it was substantially lower and ranged from a maximum of 573,27 euros, for gross earnings lower than euros, to a minimum of 51,65 euros, 8

9 for gross earnings higher than euros. Also the tax credit for a dependent spouse decreases with liable income. To be eligible for this type of credit, the spouse must have a personal income lower than a very modest threshold 7 In 2002 it varied from a maximum of 546,18 euros for income lower than ,71 euros to a minimum of 422,23 euros for income higher than ,69 euros. Finally, the third main form of tax credit is the one for dependent children: it depends negatively on family income and positively on the number of children within the family. The amount of credit can be shared by both parents if both have taxable income. In 2002 it varied from a maximum of 546,18 to a minimum of 285,08 euros. An additional xed tax credit of 123,95 euros was given for each baby younger than 3 years. The Italian scal system also includes two major social transfers linked to the family income and structure (table 2.3). The "family allowance" is given to employed or retired individuals that have at least one child younger than 18. "The family allowance for young children", instead, is given to families that have at least three children younger than 18, irrespective of the claimant employment status. The transfer amount and the income level for eligibility increase according to the number of underage children and decrease with the family income. Both requirements are systematically higher for single parents than for couples. 9

10 III The Data The present empirical analysis is carried out using the 2002 Bank of Italy Survey of Household Income and Wealth (SHIW). SHIW provides detailed information on a representative sample of the Italian population including micro data on socioeconomic characteristics, labour and non-labour income and wealth of 8011 Italian families (21148 individuals). Since we focus on female labor supply, we selected a sub-sample of women between the age of 18 and 55, either employed or not. Individuals still in education, self-employed or retired were excluded. The nal selected sample is made by 4227 women divided into two sub-groups: 2919 married women 8 and 1308 single women, 388 of which living on their own and 920 living within the parental household. Descriptive statistics for the two sub-samples are shown in table 3.1. TABLE 3.1 HERE Married women are on average older (by 10 years) and less educated 9 than single women; 83,93 percent of them have at least one child, with an average of 1,8 each, against the 12,38 percent of single women 10, with an average of 1,5 each. 13,98 percent of married women have babies (children younger than 3), while very few singles, among those who have children, have babies (less than 10

11 5 percent). Both married and single women are more likely to live in a house they or their family own than to live in rented accommodation. More than 70 percent of single women still live with their original family. Married women are less likely to participate in the labour market than single women: less than 50 percent of married women work while more that 65 percent of single women are employed. By dividing the participation rate for geographical areas, we observe that participation in the labour market for married women is higher than 50 per cent in both northern (61 percent) and central areas 11 (52 percent), but the overall participation rate is forced down by the very low rate in southern regions (only 26 percent). A similar path exists in the sub-sample of single women where participation rates are very high both in northern and central areas (respectively 85 percent and 76 percent) and under 50 percent only in southern regions (40 percent); participation rates are, in any case, always higher than the corresponding ones in the married sub-sample. Married women on average earn more than single women (slightly less than 8 euro per hour against slightly more than 7 euro per hour) and work a couple of hours less per week. Everywhere apart from in southern regions, where the ratios are pretty similar, part-time work, as shown in table 3.2, is more common among married than among single women. 11

12 IV The Labour Supply Model The standard assumption in neoclassical models of labor supply is that individuals can decide to work a number of hours equal to each positive real number. However, in reality, individuals, most of the time, can choose between part-time or full-time jobs with a predetermined number of working hours. To account for this hours constraint, we use the discrete choice structural labor supply model developed by Van Soest In this model, each family can choose among L alternatives in the choice set made by income and working hours combinations f(y l ; h ml ; h fl ) ; l = 1; 2; :::; Lg ; where h ml and h fl are working hours per week of husband and wife. Possible working hours are multiple of some xed interval length IL, creating a discrete number of possible alternatives instead of a continuum as in neoclassical labour supply models. Since the focus of this paper is female labour supply, we will treat husband labour supply as xed at the observed values 12, reducing the family choice set to combinations of family income and wife s working hours. We denote by y l family s after tax income associated to the l alternative, made up of husband s earnings, wife s earnings and family unearned income such as capital income and social transfers. In the model what matters is how the family budget set is determined by the wife s working decisions, not its shape. Therefore, nonlinear and large non-convex portions caused by the presence of mean-tested social transfers are easily handled in this type of approach. We use a translog speci cation of the direct utility function: V (v q ) = v 0 Av + b 0 v (4) where v q = (log y q ; log h qf ) 0 is the vector of log commodities of the family q and A, a 2x2 matrix with entries a ij (i; j = 1; 2); and b, a 1x2 vector with 12

13 entries b i (i = 1; 2); are parameters to be estimated. Preferences variations across families due to observed characteristics can be incorporated through parameters in the following way: i = X k b ik z k ; i = 1; 2 and ij = X k a ijk z k i; j = 1; 2 (5) The zk 0 s re ect family characteristics such as family composition, wife s age, where the family lives, and include a constant term. In the empirical analysis, to reduce computational burden, A will be assumed to be constant across families and Z q will be a 1x12 vector. The nal form of each family s direct utility function is: V (log y q ; log h qf ) = 1 log y q + 2 log h qf + 11 (log y q ) (log h qf ) 2 + ( ) log y q log h qf (6) Family q disposable income corresponding to the l choice, y ql, could be expressed as a function T of family gross income and socio-demographic characteristics: y ql = T (w q h fl ; t ql ; I q; Z q ) (7) where w q h fl are woman s earnings, computed using the hourly gross wage rate w 13 q, I q is the exogenous income, made up of household unearned income and husband s earnings, in the case of married women, or parents earnings, in the case of single women living within the parental household, and t ql are the social transfers received by the family: 13

14 The empirical analysis consists of estimating preferences directly as revealed by individual choices, rather than through the speci cation of the labour supply function. Household q chooses one among L alternatives in the choice set. The utility the household can derive from each alternative l is given by: U ql = V (h fl ; y ql ; Z q ) + ql (8) where V () is the utility function de ned in equation 6 and ql is an error term 14 assumed to be identically and independently distributed across alternatives and across families according to a type I-extreme value distribution. Under this assumption, McFadden 1974 proved that the probability that alternative n is chosen by household q is given by: P r qn = Pr(U qn > U ql ; 8l 2 L) = exp V (h n; y qn ; Z q ) LX exp V (h l ; y ql ; Z q ) l=1 (9) that leads to the estimation of a conditional logit model. Italy, as many countries, shows a concentration of people around the parttime, full-time and non-working alternatives. The above outlined model is not able to replicate these peaks. Therefore, to improve the t of the model, it is common practice to add either dummies (as in Van Soest 1995), that can re ects quantity constraints on the demand side, or a xed costs variable (as in Bargain and Orsini 2006), that represents the direct and indirect costs an individual has to cover to work (like transport costs and child-caring costs). We use the xed costs approach modelling them as a one-o weekly cost directly subtracted from net income for any choice that involves paid work. They enter in the utility comparison for each individual in their work - non work choice in the following form: 14

15 Density Density F = X F (10) Since we assume non stochastic xed costs,they do not modify the likelihood function. The functional form of the utility function, instead, becomes: 8 >< V (h l ; y ql ; Z q ) + ql if h l = 0 U ql = >: V (h l ; y ql F; Z q ) + ql if h l > 0 (11) In our sample, the working hours reported by individuals range practically all integers from 0 to 70. It is, then, necessary to use a grouping rule that maps the declared hours into a discrete number of possibilities Women in couple Single women Figure 4.1: hours worked by married women Figure 4.2: hours worked by single women Gigure 4.1 and 4.2 show the distribution of working hours in the selected sample for single and married women, while gure 4.3 presents the number of hours worked by type of contract 15. It is evident a strong concentration around the full-time (30, 35 and 40 hours) peaks and a minor concentration around the part-time peak (20 hours). 15

16 Density Graphs by partime oretotd Figure 5.3: hours worked by type of contract We use three di erent grouping rules constructed using two interval lengths (IL = 20 and IL = 10) and three set sizes (L = 3; L = 5, L = 6) 16 to test if our results are robust to di erent choice sets. Finally, a well known problem in the labour supply literature is that wages are observed only for those actually working. Therefore, it is necessary to impute a wage to individuals who are currently out of work taking into account the bias linked to participation decisions. A popular solution is to use the Heckman correction Heckman Strictly speaking, Heckman corrected wages might induce a correlation between the income and the utility stochastic component. Due to the selectivity problem, individuals with a large positive stochastic component in the wage equation are more likely to be observed in employment, given the observed variables; therefore, the wage stochastic component and the utility stochastic component become correlated. To x this problem, di erent solutions have been adopted, all questionable to a certain degree: a) to use only the systematic component of the wage equation for everyone; b) to use the observed wages for employed individuals and the predicted wages for not employed; c) to use the predicted wages for all individuals. A more sophisticated procedure, that avoids this 16

17 correlation, simultaneously estimates the wage equation and the utility function or, alternatively, integrates the likelihood with respect to the distribution of the wage stochastic component. We alternatively use all the four possible wages, the three above mentioned possibilities based on the Heckman correction and a fourth possibility constructed using a numerical procedure to approximate the integration of the likelihood with respect to the wage stochastic component 17, to test whether the Heckman procedure induce a bias in the labor supply coe cients. The results of the wage estimation are presented in the appendix. V Empirical Results Before looking at the empirical results, it is important to stress that they have to be interpreted with caution as preferences of individuals in a static environment, because this model does not explicitly take into account demand-side factors, as rationing in disposable working hours, and factors that might in uence individuals behaviors and preferences in a dynamic perspective (Bargain and Orsini 2006). We estimate utility parameter, as revealed by actual working choices, using maximum likelihood. We allow xed costs to vary according to the number of hours worked (part-time, full-time or over-time). Moreover, we tried to interact xed costs with some observable factors that in principle should raise or lower their impact on individual choices (as the presence of young children within the household or the region of residence) but all coe cients di erent from the main one proved to be statistically not di erent from zero, therefore we do not include them in our simulation framework. The grouping rule based on the type of contract (IL = 20; L = 3) was able to t the data, in terms of participation decisions, quite well without the introduction of the xed costs 17

18 variables, therefore we use them only in the other two cases based on the declared number of working hours (IL = 10; L = 5 or L = 6). For single women living with parents two possible types of family income have been considered, to investigate if parents working situation a ects cohabiting children s working decisions. The rst possibility includes daughter s earnings, mother s earnings, father s earnings and family unearned income, while the second type is made up of only daughter s earnings and family unearned income. Results for the two types of income are much the same, suggesting that the parents earnings do not have a direct e ect on daughter s working decisions. In the following, we present the results obtained using the income of the whole family. In our original sample, 51% of married women and 8% of single women declared to be housewives. In our estimates, we want to control for the fact that these women might have a strong preference towards the non working status that can determine their participation decisions in a way not related to economic reasons. We, then, include in the utility derived from working a dummy variable equal to 1 if in the original dataset the woman is a housewife. Finally, we want to control for the fact that poor families might have some unobservable characteristics, like for example a poor social network, that might in uence their working choices. We create a dummy equal to 1 if, in the observed data, the woman s family has an income below a certain threshold and we include it in the utility derived both from working and from income. In the following we show the estimates obtained using the experimented minimum income threshold 18 as benchmark for the construction of the dummy variable. Table 5.1 and 5.2 show the results for married and single women using two out of three grouping rules 19 (IL = 20, L = 3 and IL = 10, L = 5) and three types of wages 20. We omit the IL = 10, L = 6 because the coe cients and the 18

19 psedudo-r2 for L = 5 and L = 6 for both married and single women are the same under all possible types of wages, implying that explicitly modelling the overwork possibility does not improve the ability of the model to replicate real choices. For both married and single women, taste parameters associated with working hours are more signi cant than those associated with income independently of the type of wage used. Signi cant coe cients have the same sign under all possible wages. Hours coe cients are larger with IL = 10 than with IL = 20, but this change in magnitude is mainly due to the presence of xed costs. In fact, hour coe cients of the IL = 10 models without xed costs are basically equal to the IL = 20 case 21. Fixed costs are always strongly signi cant and vary according to the number of hours worked. In particular, they decrease as the number of hours increases, though the change is small with respect to the change in the number of hours. This result suggests that only a small fraction of the cost of working is related to the number of hours worked; the largest part is a sort of sunk cost related only to the participation-non participation dichotomy. In the case of married women, the main hours coe cient is always negative, as expected since we use working hours and not leisure time 22, and strongly signi cant. The number of children within the household has a positive impact on the utility of working (it weakens the disutility derived from participation) and a negative impact on the utility derived from income (signi cant in most but not all of the tried speci cations). The latter e ect could be related to the fact that the higher the number of children within the household, the higher the number of individuals competing for the same economic resources. A negative, then reinforcing, e ect on the utility linked to labor participation is associated both with living in central and in southern regions, the latter being stronger 19

20 than the former. Age shows the usual concave pattern but the coe cients are not statistically di erent from zero. Under random wages, home ownership reinforces the disutility derived from working. A possible explanation could derive from the need of a second source of labour income to support or simply to easier the refund of the loan most of the time associated with house purchases. The housewife coe cient is always negative, increasing the disutility derived from working as expected, but it is not always signi cant 23. Nevertheless, the inclusion of this variable increases signi cantly the pseudo-r2 under all possibilities, both for married and single women. To be below the threshold has always a negative impact on the utility derived from income, but it is signi cant only using corrected Heckman wages. It also has most of the time a negative, then reinforcing, e ect on the utility derived from working, but it is signi cant only when no Heckman correction is considered. These two coe cients are the only ones not robust to the di erent speci cations. TABLE 5.1 HERE The coe cients for single women are similar to those already commented for married women. Cohabiting with parents increases the disutility derived from working. A possible interpretation could be related to the fact that single cohabiting women face lower wages 24, implying that they face less attractive job o ers. Another possibility is that single women that live with parents most of the time do unpaid housework and, therefore, are less likely to get a paid job. The e ect of home ownership is insigni cant on the utility derived from working but it has a negative e ect on the utility derived from income. Also in the case of single women, the housewife coe cients is always negative but not always signi cant. Both coe cients related to the income threshold are negative and, di erently from married women, they are signi cantly di erent from zero under all possible wages. 20

21 TABLE 5.2 HERE An important characteristic of the model, to get reliable results on the simulation exercise, is its ability to replicate the actual data in terms of working hours frequencies. To verify the ability of our di erent speci cations to t the actual sample, we report in tables 5.3 the observed and the average predicted frequencies. All possible combinations of wages and choice sets t the observed frequencies of married women very well. Random wages tend to slightly overpredict full-time work and to underpredict the non working status. All speci cations are also able to replicate real working decisions of single women cohabiting with their parents. The speci cations that use IL = 10 tend to overestimate the intermediate solutions (10, 20 and 30 hours) and to underestimate the non working status. As in the case of married women, when random wages are used we get the worst scenario. Full-time work is strongly overpredicted while all the other hours possibilities are underpredicted. Finally, our model seems to fail in representing working decisions of single women living alone. In fact, none of our speci cation is able to replicate exactly real frequencies. All combinations of wages and choice sets overestimate fulltime (40 hours) and over-time work and underestimate all the other possibilities. Heckman corrected wages generate frequencies more similar to the observed ones. TABLE 5.3 HERE VI Policy Design and Simulation Results In this section, we use our estimated labor supply coe cients to simulate the e ect, on female labor decisions, of the introduction in the Italian welfare system 21

22 of a minimum income policy shaped on the scheme carried out from 1998 to The reddito minimo d inserimento was a participation basic income made up of a nancial part, established by the central government and managed by the local governments involved in the project, and of a participation scheme designed and managed entirely by local governments. We are only able to simulate the e ects of the nancial part of the program. The bene t scheme was mainly characterized by three elements: the family reference income, the threshold level and the labor earnings inclusion mechanism. The reference family income for eligibility was made up of all family members taxable income plus one fth of household nancial capital and one fth of household real capital, calculated using ISE rules. The eligibility income threshold was equal to 282 euros per individual and it was adapted to family size and characteristics using, again, ISE scale. Finally, only 75% of labor earnings of each family member was included in the family reference income. We test the existence and magnitude of the labor disincentive e ect using di erent eligibility thresholds 25 and di erent levels of earnings inclusions On on side, the lower the income threshold the more stringent the income constraint should be and the higher the probability of loosing the transfer even if the individual has a very poorly paid job. Therefore, the disincentive e ect should be weakened by an increase in the threshold level. On the other side, due to the income e ect, the higher the eligibility threshold, the higher the transfer and, as a consequence, the higher the incentive individuals face not to work. Higher eligibility thresholds should, then, reinforce the disincentive e ect. Which e ect prevails is not a priori certain. The earnings inclusion mechanism, instead, has a clear relation with the disincentive e ect. The higher the level of labor earnings included in the family reference income, the stronger the e ect should be. 22

23 We try ve di erent income thresholds, set respectively equal to the experimented one, the minimum pension level, the absolute poverty line and the relative (full and 80%) poverty line 26, and ve di erent levels of earnings inclusion, using from 75% up to 100% of each individual s labour earnings. Table 6.1 reports the ratio of individuals below the di erent income thresholds 27. Single women living alone are the group that su ers the most in terms of income. In fact, it shows the highest percentage of individuals below the threshold under all ve possibilities. They are also the group that bene ts the less from the actual welfare system. Since family allowances and allowances for young children, based on the 2002 Italian tax-bene t system described in section 2, were included in family unearned income used to estimate the model in the previous section, we also include them in the status quo simulation. In the other scenarios they are replaced by the di erent types of minimum income transfers. We also include a benchmark scenario where no social transfer is available. Tax credits for dependent spouses and children have been maintained in all the simulations. Table shows the simulation results based on (IL = 20; L = 3) and type I wages estimates, a model with a small, then rigid, choice set but a high ability to replicate the observed working frequencies. TABLE 6.2 HERE For married women, the baseline case (the one without social transfers) is 23

24 the worst in terms of labour participation rates, having the highest ratio of individual in the non working status. This fact supports the idea that actual social transfers are able to weaken the economic constraints that in poor couples prevent women to participate in the labour market. When we look at the minimum income scenarios, a disincentive e ect comes out but the size is very limited under all possible transfer schemes and it decreases as the threshold level increases. The disincentive e ect concerns more full-time work than parttime work. Earnings inclusion does not have any e ect on married women labor supply decisions. Results for single women are quite di erent. For single women living with parents, social transfers are linked to the whole family situation, including the parents. What is, then, relevant is if by taking up a paid job, the daughter will cause her family to lose the transfer it was entitled to receive. In this case, we never observe a decrease in labor participation. Participation is positively correlated with the income threshold and its increase is less pronounced when higher ratio of labor earnings are included in the reference income. The same results hold for single women living alone. Minimum income transfers seem, then, to allow single women to work or to work more. This result could be in uenced by the fact that existing social transfers and simulated minimum income policies reach di erent targets. Existing social bene ts are mainly for families, implying that single individuals have no access to them, while minimum income policies are designed for individuals and adapted to family composition through the ISE system. We run simulations using all possible combinations of choice sets and types of wages. They all lead to the results we described above. A labor disincentive e ect exists only in the case of married women, but its size is very limited. The income threshold has always a positive e ect on female labor supply decisions, 24

25 while the earnings inclusion mechanism has basically no e ect. Our results, in fact, hold under all possible earnings inclusion levels, even 100%. This suggests that individuals, for their participation decisions, consider mainly the level of the income threshold and do not take into account the fact that part of their earnings might be excluded from the reference income. VII Conclusions Minimum income policies are often seen as an e ective instrument to ght poverty and social exclusion. Their main weakness relies on the theoretical disincentive e ect on labour market participation they may cause at the bottom end of the income distribution. The problem is that individuals with low wages and not so attractive job perspectives could shortsightedly nd it more convenient to remain on purpose out of the labour market or even to become unemployed in order to be included in the welfare programme. In the long run, of course, this is a highly undesirable e ect. In this work we try to test the existence and the magnitude of this labor disincentive e ect by estimating a discrete choice structural labor supply model on Italian data and, then, by simulating the e ect of di erent minimum income schemes. Di erently from the existing literature, we focus our attention on the e ect linked to the public transfer, isolating it from the e ects due to changes in the tax structure. Our results suggest that it is not at all obvious that minimum income policies have such a disincentive impact on employment, at least in the case of Italian women. We considered di erent groups of women (married women, singles living with parents and singles living alone) and di erent combinations of income eligibility criteria and levels of labour earnings exemption. Theoretically, the level of the income threshold has both a positive and a negative e ect on labor 25

26 decisions, while the level of earnings inclusion should has a negative e ect. Our results show that the mechanism of labour earnings inclusion, studied explicitly to avoid the disincentive e ect, seems to play no role in female participation decisions. The level of the income threshold, instead, matters and has an overall positive e ect on participation decisions: the higher the level, the weaker the disincentive e ect. Moreover, this e ect comes out only in the case of married women, but it tends to weaken the higher the threshold used. Single women, both living with parents and women living alone, always increase their labor market participation when they receive a minimum income transfer. Our results are in line with those of other existing studies (Aaberge et al. 2005, Bargain and Orsini 2006, Blundell et al. 2000) and suggest that single and married women respond di erently, in terms of labor decisions, to policy measures. They also suggest that in Italy the actual welfare system seems to bene t only married women, while a general transfer would reach single women too, in particular those living alone. Since they are also the group more likely to experience poverty, our results suggest that minimum income transfers would allow them to decide more freely about their working hours by relaxing the economic contraints they face. Further work is needed to verify if these results hold also with di erent tax structures and welfare systems. Acknowledgments I thank Daniela Del Boca for her outstanding supervision and Ugo Colombino and Giuseppe Bertola for their helpful comments. I also thank Massimo Baldini for providing me with gross earnigs generated by his simulation program Mapp98. I gratefully acknowledge the nancial and logistic support of Fondazione CRT and Collegio Carlo Alberto. 26

27 Notes 1 A 1992 European recommendation suggest that European governments should have some sort of universal basic income mechanism. 2 O course, if t 1 = 1 we go back to the basic NIT structure with a 100% marginal tax rate around the threshold. 3 The evaluation process was, for the rst time in Italy, assigned to an independent institution, but the results never became public. 4 In 2002 money value. 5 The ISE (Indicatore della situazione economica) scale allows us to calculate equivalent income for families with di erent characteristics. Starting from a weight equal to 1 for a single member family, it increases by 0.35 for every additional member and by an additional 0.2 for particular situations such as single parents, couples where both parents work and disabled children. 6 With an upper limit of 750 working hours in 1998, now extended, after which all earnings enter into the relevant income. 7 In 2002 it was equal to 2.840,51 euros, meaning basically that he or she has not work on a regular base. 8 The term married refers to both spouses and cohabiting couples. 9 Low level of education = less or equal to compulsory education; mid level of education = high school or equivalent; high level of education = graduation or higher. 10 In the SHIW survey, individuals reported as son/ daughter in the original family structure provide no information on their own family (spouse and children). Therefore, married daughters have been excluded from the sample while the number of children for single daughters has been set equal to zero by hypothesis. 11 The northern area includes Valle d Aosta, Piemonte, Liguria, Lombardia, Trentino-Alto Adige, Friuli-Venezia Giulia, Veneto and Emilia-Romagna; the central area includes Toscana, Umbria, Marche and Lazio; the southern area includes Abruzzo, Molise, Campania, Puglia, Basilicata, Calabria, Sicilia and Sardegna ,24% of husbands in our sample work. If we exclude retired husbands, the ratio goes up to more than 94%. 13 Which is assumed not to vary across alternatives. 14 Error terms can be interpreted as unobserved alternative s speci c utility components or errors in perception of the alternative s utility. 15 1= full-time, 2=part-time 27

28 16 First rule: L = 3 and IL = 20. Individuals are assigned to each alternative looking at the type of contract they have. h = 20 if the individual works part-time and h = 40 if the individual works full-time. Second rule: L = 5 and IL = 10. Individuals are assigned to each alternative using the declared working hours and the following classes: h = 0 if h 5 or missing, h = 10 if 5 < h 15, h = 20 if 15 < h 25, h = 30 if 25 < h 35 and h = 40 if h > 35. Third choice: L = 6 and IL = 10. Individuals are assigned to each alternative using the declared working hours and the following classes: h = 0 if h 5 or missing, h = 10 if 5 < h 15, h = 20 if 15 < h 25, h = 30 if 25 < h 35; h = 40 if 35 < h 45 and h = 50 if h > 45. Using this rule, we explicitly model overwork decisions. The after-tax income y l is computed using the imputed working hours. 17 See Van Soest We create dummies for all ve possible income thresholds showed in the next section and we always get the same estimation results. 19 Tables showing all results are available upon request. 20 Type I = only the systematic part for all individuals; type II = observed wage for workers and predicted wage for non-workers; type III = predicted wage for all individuals; type IV = random wage for all individuals. In the last case, standard errors have been bootstrapped. 21 Results are not shown but available upon request. 22 We do not have time use data. As a consequence, we cannot divide non-working time into leisure and time devoted to activities (like housekeeping) that could potentially produce disutility. Therefore, we prefer to use working hours. 23 Only in the IL = 20, type I wage and under random wages. 24 See wage estimation results shown in the appendix. 25 Of course di erent thresholds imply di erent total costs for the national scal system, but an analysis of the scal sustainability of the di erent possibilities is beyond the scope of this paper. Our focus is simply to test the impact of di erent thresholds on participation rates. 26 Absolute poverty and relative poverty as calculated for year 2002 by Istat (see for more information). 27 As reference income, we consider the family income as computed using observed data. 28 A=no social transfer; B=minimum income with experimented income threshold; C=minimum income with minimum pension as income threshold; D=minimum income with absolute poverty line as income threshold; E=minimum income with 80% of relative poverty line as income threshold; F=minimum income with relative poverty line as income threshold. 28

29 References Aaberge, R., U. Colombino, and S. Strøm (1999). Labour supply in italy: An empirical analysis of joint household decisions, with taxes and quantity constraints. Journal of Applied Econometrics 14, Aaberge, R., U. Colombino, and S. Strøm (2004). Do more equal slices shrink the cake? an empirical investigation of tax-transfer reform proposals in italy. Journal of Population Economics 17, Aaberge, R., U. Colombino, and S. Strøm (2005). Taxes, transfers, labor supply and household welfare. In D. D. B. T. Boeri and C. Pissarides (Eds.), Women at Work. An Economic Perspective. Oxford University Press. Bargain, O. and K. Orsini (2006). In-work policies in europe: Killing two birds with one stone? Labor Economics 13 (6), Blundell, R., A. Duncan, J. McCrae, and C. Meghir (2000). The labor market impact of the working families tax credit. Fiscal Studies 21 (1), Colombino, U. and D. Del Boca (1990). The e ect of taxes on labor supply in italy. The Journal of Human Resources 25, Dickens, W. and S. Lundberg (1993). Hours restrictions and labor supply. International Economic Review 34, Fortin, B., M. Truchon, and L. Beauséjour (1993). On reforming the welfare system. workfare meets the negative income tax. Journal of Public Economics 51 (2), Gouveia, M. and C. Rodrigues (2002). The impact of a guaranteed minimum income program in portugal. Public Finance and Management 2 (2). Gurgand, M. and D. Margolis (2005). Does work pay in france? monetary incentives and the guaranteed minimum income. IZA discussion paper

30 Heckman, J. (1979). Sample selection bias as speci cation error. Econometrica 47, pp Laroque, G. and B. Salanie (2002). Labour market institutions and employment in france. Journal of Applied Econometrics 17, McFadden, D. (1974). Conditional logit analysis of qualitative choice behaviour. In P. Zarembka (Ed.), Frontiers of Econometrics. New York: Academic Press. Rodrigues, C. (2003). The redistributive impact of guaranteed minimum income programme in portugal. Working paper. Sacchi, S. (2005). Reddito minimo e politiche di contrasto alla povertà in italia. URGE working paper n.1/2005. Targetti Lenti, R. (2000). Reddito di cittadinanza e minimo vitale. Società Italiana di Economia Pubblica working paper. Van Parijs, P. and Y. Vanderborght (2006). Il reddito minimo universale. Egea. Van Soest, A. (1995). Structural models of family labor supply: A discrete choice approach. The Journal of Human Resources 30 (1), A Wage Estimates Table A.1 and A.2 present the results of the Heckman procedure for gross and net wages, separately for married and single women..since SHIW reports only information on net incomes it was necessary to recover gross wages using an ad hoc microsimulation program 29. The included variables could be divided into the following main categories : individual characteristics: age (divided by 10), age squared (divided by 30

31 100) and educational level for both the selection and the main process; family characteristics: the number of children and the number of babies (children younger than 3) within the household, the ownership or the rent of the house, the area of residence (North, Centre or South), the presence of grandparents in the household and the family unearned income (divided by 100), and in the case of single women, a dummy equals to one when she lives with her parents and zero otherwise; in the case of married women, husband characteristics related to his labour income. We do not use directly the husband net labour income because it is likely to be correlated to the wife s wage, due to its dependence on tax credits for children and family arrangement shared by the spouses. To avoid this problem, the husband s earnings are represented by the husband s level of education, type of job and working sector 30. Looking at the results for married women, all variables in the selection process are statistically signi cant with the exception of house rental, having a baby and family unearned income. Age seems not to have a direct e ect on wages but it has a strong positive e ect on participation, decreasing with woman s age. Living in central and especially in southern regions lowers the probability to work. Women that live in central Italy also have lower wages. The educational level has an impact on both processes: the higher the educational level, the higher the probability that a woman will work and the higher the wage she will get. The direct e ect on wages is stronger than the one on participation, especially in the case of gross wages. The greater gain in term of earnings is generated by reaching graduation compared to all other educational attainments, while the di erence between high school and low level of education is still statistically signi cant but much smaller. Having children lowers wife s participation rate independently from their age. Home ownership has a posi- 31

32 tive e ect on participation; a possible explanation could derive from the need of a second source of labour income to support or simply to easier the refund of the loan most of the time associated with house purchases. The husband s education has a positive e ect on the wife s participation; this e ect could be related to the assortative mating phenomenon. Namely, men with higher education are likely to be married with women that also have a high education and that, consequently, are more likely to work and to get high wages. Finally, the husband s working position 31 has a negative impact on wife s participation. TABLE A.1 HERE The coe cients for single women are similar to those already commented for married women. An interesting point is that single women living with parents earn signi cantly less than the others. A possible explanation for this coe cient could be that, by living with their parents, these women face lower living costs and, therefore, are able to accept jobs with lower wages (at least initially). Single women that cohabit with parents also participate less in the labour market. A possible explanation is that, most of the time, they are engaged in unpaid work within the household and, therefore, are less likely to get an outside paid job, unless strongly motivated. TABLE A.1 HERE 32

33 33

34 34

35 35

36 36

37 37